Public FR3 events show no recall rescue from closed-loop stimulation in 17 analyzed subjects

We analyzed 9,384 words from 32 FR3 sessions in 17 of 18 participants in OpenNeuro ds005557 v1.0.0. Public behavioral events do not include the classifier output, so the classifier-matched test of Ezzyat et al. (2018) cannot be repeated. Stimulation events were assigned to the preceding word by onset time; 48% of words on stimulation lists triggered stimulation. Triggered words had recall 6.3 percentage points below non-triggered words on stimulation lists (95% CI -9.3 to -3.2; p=0.002) and 3.7 points below the same-position no-stimulation baseline (95% CI -5.6 to -1.6; p=0.004). These item-level contrasts combine classifier selection and stimulation and are not causal. All stimulation-list words differed from baseline by -0.5 points (95% CI -2.3 to +1.2; p=0.57). A post hoc estimate was -0.9 points per triggered word (95% CI -5.1 to +3.1; p=0.68), and is causal only if lists were randomly assigned; it also includes effects on neighbouring words. The public events show no evidence of recall rescue, within these limits.

closed-loop stimulation · free recall · memory · intracranial stimulation · FR3 · OpenNeuro · item-level analysis